Performance enhancements and visualization for RSQSim earthquake simulator

Dmitry Pekurovsky, Amit Chourasia, Keith B. Richards-Dinger, Bruce E. Shaw, James H. Dieterich, & Yifeng Cui

Submitted August 14, 2017, SCEC Contribution #7596, 2017 SCEC Annual Meeting Poster #011 (PDF)

We report on progress of performance tuning and visualization for earthquake simulator code RSQSim. We have studied performance of the code in detail on supercomputers such as Blue Waters (NCSA/UIUC), Stampede (TACC) and Mira (Argonne). We have continued to improve scalability efficiency of the code, mainly by reducing load imbalance. This is achieved by introducing a new feature of the code which uses an added physical constraint that allows a large reduction in the number of times when an iterative solve is called. This saves on calculation time, as well as makes the code more load-balanced due to less variable numbers of calculation steps. This feature alone leads to up to 33% better performance and substantially improved scalability. Furthermore, we have improved the scaling of the initialization portion of the code and created a Github repository, for the first time, for the SCEC CISM project.

We also report final results from visualization work. RSQSim generates data catalog of millions of seismic events for thousands of years and their associated properties such as changes in slip, stress, etc. Exploring this data is of vital importance to validate the simulator as well as to identify features of interest such as quake time histories, conduct analyses such as calculating mean recurrence interval of events on each fault section. We have created a prototype web based tool for exploring this data interactively. In course of visualization work we recognized the need to transform the data from current form into a more scalable and manageable format. We also summarize data management findings and discuss how the changes impact analysis and its future potential. Interactive visualization is accessible at http://vis.sdsc.edu:5555


Key Words
Earthquake Simulation, High Performace Computing, Supercomputers, Performance, Scalability, Software

Citation
Pekurovsky, D., Chourasia, A., Richards-Dinger, K. B., Shaw, B. E., Dieterich, J. H., & Cui, Y. (2017, 08). Performance enhancements and visualization for RSQSim earthquake simulator. Poster Presentation at 2017 SCEC Annual Meeting.


Related Projects & Working Groups
Collaboratory for Interseismic Simulation and Modeling (CISM)